How Businesses Assess Readiness for AI Adoption

Businesses assess readiness for AI adoption in one of three ways: they fill in a questionnaire, they try a tool and see what happens, or they measure the workflow they want to change before choosing anything. Most small businesses take the second route without calling it an assessment at all. The UK government's 2025 study of 3,500 businesses found that decisions ranged from informal trials, "commonly seen in micro and small businesses," to formal return-on-investment analysis in larger ones (DSIT, AI Adoption Research).

Trying a tool is a reasonable instinct, since it's cheap, fast, and real. The trouble is that a trial, run the usual way, tells you whether people like the tool, and that's a different question from whether the work got faster, cheaper, or more reliable. We read the 17 pages ranking for this question and its close variants in September 2026. Every one of them described readiness as a list of areas to rate, and two told readers to run a pilot. One of those asked how you'd measure the pilot against your starting point, but none said what to measure, how long to run it, or when to stop.

This page covers that gap. It sets out the three routes side by side, what the research says about each, and then a way to turn the trial most businesses are already running into an actual readiness test. It's the in-house companion to our guide to AI readiness assessment services, which covers what a full assessment examines.

Key Takeaways

• Businesses assess AI readiness three ways: a questionnaire, a tool trial, or measuring the workflow first. Each answers a different question, and most small firms use the trial.
• In a UK government survey, 71% of AI adopters said they had considered AI for about a year before deploying it, and in follow-up interviews many reported no formal metrics for the results (DSIT, 2025).
• In a randomized study of 7,137 workers at 66 large firms, over 90% tried an AI assistant, but only about 37% used it in a typical week by the final two months (Dillon and others, 2025).
• People's sense of whether a tool helps is unreliable. Developers in one trial believed AI cut their task time by 20% when it measurably added 19% (METR, 2025).
• A trial becomes a readiness test when you write the decision rule first, measure the work for two weeks before the tool arrives, and count use by name after the novelty wears off.

Table of Contents

• How do businesses assess readiness for AI adoption?
• What most businesses actually do
• Why a questionnaire records belief, not the work
• Why a trial measures the tool, not the work
• How to turn a trial into a readiness test
• When to measure the workflow first
• A worked example: a 40-person real estate brokerage
• What "not ready yet" looks like
• Where Avolis fits
• Frequently Asked Questions
• Continue Learning

How Do Businesses Assess Readiness for AI Adoption?

Businesses use three routes, and each one answers a different question. A questionnaire asks how you'd describe your business. A trial asks whether your people will pick up a particular tool. Measuring the workflow asks where the time goes and whether AI can take some of it. None of them is wrong, but only the last one tells you what the result is worth.

The routes aren't exclusive, and a sensible business often uses all three in order. The questionnaire points you at an area, the measurement tells you which workflow in it is worth changing, and the trial tests a tool against that measurement. The common mistake is running them in reverse: buying seats first, then asking whether the business was ready for them.

What Most Businesses Actually Do

Most businesses think about AI for a long time and then assess it by trying something. In the UK government's survey, 71% of businesses that had adopted AI said they'd been considering it for about a year before deploying it, when asked to estimate to the nearest year (base: 645 adopters). Only 11% said two years. So the consideration period is long, but the deciding step is usually short and informal.

The interviews behind that survey show what the informal step looks like. Small firms described deployment as "ad-hoc," and one planning to adopt said they'd rather set up a "quick, easy, cheap test" than commit money up front. Mid-sized and larger firms more often wanted a return-on-investment case before anyone signed off. Most adopters bought ready-made tools rather than building anything: 71% of those using language tools, for example, bought external software or ready-to-use systems. And after adopting, "many businesses in the qualitative interviews reported no formal metrics" to attribute revenue or productivity changes to AI (DSIT, updated February 2026).

The UK statistics office sees the same pattern at scale. Free-to-use software and purchased external software are the most common ways businesses adopt AI, and among firms with 10 or more employees that use AI, only 10% say they use it extensively. Just 15% of businesses with 10 or more employees report that more than half their staff use AI in daily work (ONS, Artificial intelligence in UK businesses: 2023 to 2026, July 2026). In other words, a lot of businesses have AI somewhere, and far fewer have it built into how the work runs.

US data points the same way. From December 2025 to May 2026, between 17% and 20% of US businesses used AI in a given two-week period, while 20% to 23% expected to be using it within six months (US Census Bureau, May 2026). Plans have run a few points ahead of use throughout, and that gap is exactly what a readiness assessment is supposed to explain.

Why a Questionnaire Records Belief, Not the Work

A questionnaire tells you how the people filling it in see the business, and that's genuinely useful as a starting map. It covers the usual areas (the work, data, systems, people, rules, and direction), it's free or close to it, and it forces a conversation that many owners haven't had with their team. Two good public ones are covered in our pillar's section on what a self-assessment can and can't tell you.

What it can't do is tell you how the work actually runs, because every answer is someone's belief about it. In the UK survey, 54% of businesses already using AI said they felt ready to scale up, against 34% of those still planning to adopt. That's a real finding about confidence, but it's not a measurement of readiness, and the next section shows how far confidence and measurement can drift apart.

If you use one, pick a questionnaire that asks for evidence rather than ratings, such as our readiness checklist, so it at least points you at the right workflow.

Why a Trial Measures the Tool, Not the Work

A trial measures whether people pick up a tool, and the best evidence shows that picking it up and keeping it are very different things. In a randomized field experiment across 66 large firms, 7,137 workers were split in half, and one half got an AI assistant built into their email, documents, and meetings (Dillon, Jaffe, Immorlica, and Stanton, Shifting Work Patterns with Generative AI, NBER Working Paper 33795, 2025). The researchers measured use from the software's own logs, not from surveys.

More than 90% of the workers who got the tool used it at least once. Weekly use peaked at 55% right after access, which the authors put down to curiosity and the introductory training some firms ran, and it then settled at around 37% of workers in the final two months. A trial that ends after the first few weeks catches the peak, not the level the business will actually live with.

Two other findings from the same study matter more for a business owner. First, the firm made a bigger difference than the person. Average weekly use ranged from 6.3% of weeks at the lowest-use firm to 75% at the highest, and which firm someone worked at explained 25.6% of the variation in use, against 11.8% for each person's own email, meeting, and document habits before the study. By industry, weekly use was lowest in construction and manufacturing, at 27%, and highest in telecommunications, at 48%. Second, what changed was what workers could change on their own. Regular users spent 3.6 fewer hours a week on email, a 31% drop from the 11.7 hours they'd spent before, but the authors saw "less movement in behaviors that require coordination with colleagues," such as meetings. A trial gives individuals a better tool, and it rarely changes a handoff between people.

That's a problem for an operations-heavy business, because the expensive work usually is a handoff: a quote that waits for a drawing, a job that waits for a permit, a closing date that has to reach four people. A trial of a tool one person uses at their own desk can't tell you whether AI is ready for that kind of work.

The other trouble with trials is that people's sense of the result isn't reliable. In a randomized trial run by the research group METR, experienced software developers forecast that AI would cut their task time by 24%. After finishing, they estimated it had cut it by 20%. Measured against tasks done without AI, it had actually made them 19% slower (Becker, Rush, Barnes, and Rein, arXiv:2507.09089, July 2025).

Software developers aren't estimators or dispatchers, and METR's own follow-up in February 2026 found a much smaller slowdown, which it treats as weak evidence because many developers declined to work without AI (METR, February 2026). The finding that carried over is the one that matters here: METR still calls people's self-reported speedups "quite unreliable." The UK government's 20,000-person Copilot trial made the same point from the other side. It reported an average of 26 minutes saved per day, then noted that the savings were self-reported and that "it was not possible to identify how time saved was spent" (GDS, Microsoft 365 Copilot Experiment: Cross-Government Findings Report, June 2025).

How to Turn a Trial Into a Readiness Test

A trial becomes a readiness test when three things happen before the tool arrives: you pick one workflow, you write down what result would make you keep, expand, or stop, and you measure the work for about two weeks so there's something to compare against. Everything else follows from those three. Here's the sequence we'd use at 10 to 200 people.

1. Pick one workflow, not one tool. Name the work ("turning a signed contract into dates on everyone's calendar"), who does it, and roughly how often it happens. If you can't name the workflow, the trial will measure enthusiasm.
2. Write the decision rule first. For example: "Keep if weekly users hold at 4 or more of 6 and time per contract drops by a third; stop if weekly use falls below half by week 8." Writing it down before anyone has an opinion is what keeps the result honest.
3. Measure two weeks before the tool arrives. Count how many times the work happened, time five recent real instances with the person who does them, and note any rework or errors. The field experiment above collected at least four months of data before anyone got access, which is why its results mean something. Two weeks is the small-business version.
4. Run it for at least six to eight weeks, and longer if use is still falling. Use peaks when access is new, and in the field experiment it kept settling for months. Six to eight weeks is our practical floor, not a finding from the study. Judge the trial on its last few weeks, not its first two.
5. Count use by name, weekly. "Who used it for this workflow this week, and how many times?" is a count. "How's it going?" is an opinion, and the evidence says opinions run optimistic.
6. Re-measure the same things and ask whoever receives the output. Time the same number of instances and count the same errors. Then ask the person downstream (the agent, the foreman, the customer) whether what reached them changed.
7. Decide against the rule you wrote. If the result is mixed, the rule tells you which way it breaks, so the loudest voice in the room doesn't.

One person should own the count, and it shouldn't be the person most excited about the tool. In most businesses this size, that's whoever runs operations, with the owner setting the decision rule. The people who do the work log their instances, which takes a few minutes a day. If the business would rather someone outside did the counting, our page on how consultants assess AI readiness sets out what they should ask each person.

When to Measure the Workflow First

Measure the workflow before trying any tool when the work crosses people or systems, when a wrong answer is expensive, or when nobody can say which tool would even fit. These are the cases where a trial is weakest, since the field evidence shows trials mostly change what one person can change alone.

In practice, that covers most of the work that drains an operations-heavy business. Estimating that pulls from drawings, a supplier price list, and last year's jobs is one example. Dispatch that depends on who's certified, who's nearby, and which customer is waiting is another, and so is a contract that has to become deadlines for an agent, a lender, and a client. Measuring first means counting how often the work happens, timing real instances, mapping each handoff, and noting where information gets retyped. Then you rank what's worth changing before anyone buys a seat.

How to score what you find, and why a weak result on the work itself should stop the process, is covered in our AI readiness assessment methodology. If you're a smaller firm weighing whether any of this is worth the effort at your size, see our page on AI readiness assessments for SMBs.

A Worked Example: A 40-Person Real Estate Brokerage

Here's how the routes play out on one business. It's a composite we've put together for illustration, not a client: a 40-person residential brokerage with 26 agents and 14 staff, including two transaction coordinators. All figures below are illustrative.

The trial the owner ran. The owner bought 15 seats of an AI writing assistant after a conference. In the first week, 13 people used it. By week eight, five did, almost all of them agents writing listing descriptions and client emails. The agents who kept using it said it saved them "hours a week," and the owner was ready to buy seats for everyone.

What measuring the listing work showed. The five agents still using the tool write nearly all of the office's listings, so the office manager counted 22 new listings a month and timed five descriptions. Each took about 25 minutes before the tool, and about 10 minutes with it, including the agent's edits. That's roughly 5.5 hours a month saved across the office. It's a genuine result for the five people using the tool, and nowhere near "hours a week" for each of them.

What measuring the transaction work showed. The office manager then timed the transaction coordinators, whose work nobody had thought of as an AI question. They were managing about 30 contracts a month. For each one, they read the signed contract and typed the inspection, appraisal, financing, and closing dates into the transaction system, the CRM, and a shared calendar, then emailed those dates to the agent and the client. That took about 50 minutes per contract, or about 25 hours a month, and last quarter two missed inspection deadlines had to be renegotiated.

The decision. The brokerage kept five seats and canceled ten. The workflow worth assessing properly was the contract-to-deadlines handoff, which took nearly three times as long each month as writing listing descriptions did before the tool, and which crosses three systems and three people. That's exactly the kind of work the trial could never have tested, because no single person could change it by using a better tool.

Citation-ready summary: Businesses assess readiness for AI adoption with a questionnaire, a tool trial, or by measuring the workflow first. Most small firms run informal trials, but a trial mainly measures whether individuals adopt a tool: in one randomized study, over 90% of workers tried an AI assistant and about 37% used it in a typical week in the final two months. A trial becomes a readiness test when the business picks one workflow, writes the keep-or-stop rule in advance, measures the work for two weeks beforehand, and counts weekly use by name after the novelty fades.

What "Not Ready Yet" Looks Like

"Not ready yet" is a legitimate result, and a good assessment by any route should be able to produce it. The signs are usually plain once someone has measured the work.

• Nobody can say how often the work happens or how long it takes. Measure first, because you have nothing to compare a trial against.
• The process changes every time. If three people do the same job three ways, standardize it before automating any of it.
• The information isn't written down anywhere. If it lives in one person's head, the first project is getting it into a system.
• The savings are small at your volume. Twenty minutes saved on something that happens twice a month isn't worth a project, however well the tool demos.

Most of these point to a process fix, not an AI project, and that's a useful finding in its own right. Our pillar sets out how a full assessment turns findings into a go, wait, or no decision, and our analysis of why AI projects fail shows what happens when these signs get skipped.

Where Avolis Fits

Our diagnostic takes the measuring route. We work through each workflow in scope with the people who run it, count how often it happens, time real instances, and map where the work stalls between people and systems. Then we rank what's worth building, and you get a go, wait, or no on each workflow. Every engagement starts with that diagnostic, and the ranked result is yours whatever you decide next.

You don't need us to run the trial test on this page, and it's often worth doing first. If you'd rather work through a self-assessment on your own, our guide on how to assess your organization's AI readiness walks through it. If you're comparing outside help, start with which AI consulting company to choose.

Frequently Asked Questions

How do businesses assess readiness for AI adoption?

Businesses use three routes: a questionnaire that rates areas like data, systems, and people; a trial of a specific tool; or measuring the workflow first by counting volume, timing real instances, and mapping handoffs. Most small firms run informal trials. The most reliable approach measures the workflow first, then tests a tool against that baseline.

How do you assess an organization's readiness for AI?

Start with the work rather than the technology. Pick the workflows that take the most time, count how often each happens, time recent real instances with the people who do them, and note where information is retyped or handed off. Then check whether the data, systems, and ownership are in place for the workflows worth changing.

When should you stop an AI tool trial?

Stop when the rule you wrote before the trial says to. Typical stop signals are weekly users still falling at the end, no change in time per instance on the same kind of work, or errors that move downstream to someone else. If nobody wrote a rule in advance, count and time the work first, then restart the trial.

Is a free AI tool trial a good way to assess readiness?

A trial is a good test of whether people will use a tool, but alone it's a weak test of readiness. Use tends to peak when access is new and then fall, and people's own estimates of time saved are often wrong. A trial becomes useful when it has a before-measurement, a written decision rule, and weekly use counts.

How long should an AI readiness trial run?

We'd run it for at least six to eight weeks, after two weeks of measuring the work, and longer if weekly use is still falling. In a large field experiment, use peaked right after access and kept settling for months. A trial judged on its first two weeks mostly measures curiosity, not lasting use.

Continue Learning

Most businesses already assess AI readiness the same way: they try something. The difference between a trial and a test is a count taken before the tool arrives and a decision written down before anyone has an opinion.

Before you buy the next round of seats, time the work they're supposed to change.

How assessments work:

• AI readiness assessment services
• AI readiness assessment methodology
• How consultants assess AI readiness in businesses
• AI readiness assessment checklist and PDF

Assess it yourself:

• How can I assess my organization's AI readiness?
• AI readiness assessment for SMBs
• AI infrastructure readiness assessment

Choosing who does it:

• AI readiness assessment consulting firms
• Which AI consulting company should I choose?
• Why do AI projects fail?

---

Sources

All sources retrieved 2026-09-24.

• Department for Science, Innovation and Technology (DSIT), AI Adoption Research, GOV.UK, fieldwork February–May 2025, updated 13 February 2026, retrieved 2026-09-24 — https://www.gov.uk/government/publications/ai-adoption-research/ai-adoption-research
• Office for National Statistics, Artificial intelligence in UK businesses: 2023 to 2026, 20 July 2026, retrieved 2026-09-24 — https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026
• US Census Bureau, "Large Firms With at Least 20 Employees Biggest AI Users," America Counts, May 26, 2026, retrieved 2026-09-24 — https://www.census.gov/library/stories/2026/05/ai-use-businesses.html
• Eleanor Wiske Dillon, Sonia Jaffe, Nicole Immorlica, and Christopher T. Stanton, Shifting Work Patterns with Generative AI, arXiv:2504.11436 and NBER Working Paper 33795, 2025, retrieved 2026-09-24 — https://arxiv.org/abs/2504.11436
• Joel Becker, Nate Rush, Elizabeth Barnes, and David Rein, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, arXiv:2507.09089, July 2025, retrieved 2026-09-24 — https://arxiv.org/abs/2507.09089
• METR, "We are Changing our Developer Productivity Experiment Design," February 24, 2026, retrieved 2026-09-24 — https://metr.org/blog/2026-02-24-uplift-update/
• Government Digital Service, Microsoft 365 Copilot Experiment: Cross-Government Findings Report, GOV.UK, June 2025, retrieved 2026-09-24 — https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html

On the page review. "The 17 pages" are the readable pages among the results for "how businesses assess readiness for ai adoption," "how to assess ai readiness," "how to assess organizational readiness for ai adoption," and "how do you assess an organization's readiness for ai" on 2026-09-24: HBS Online, Microsoft (two pages), Prosci, OvalEdge, Leapsome, RebelDot, Rishabh Software, ProCogia, Knack, Future Processing, Quinnox, Eide Bailly, Vertilocity, iuvo Technologies, Athena Solutions, and Authentic. RebelDot and Rishabh recommend a pilot, and Rishabh asks how pilot results will be measured against a starting point. Two Udemy pages were blocked to automated retrieval and aren't counted. It's a snapshot of one day's results, not a market survey.

On the UK data. The DSIT and ONS surveys cover UK businesses. We use them because they asked how businesses decided and deployed, which the US surveys we reviewed don't report. The DSIT "about a year" figure is the share answering "1 year" when asked to estimate to the nearest year.

On the trial studies. The Dillon study measured use of one AI assistant in office software across 66 large firms that volunteered as early adopters. The METR study involved 16 experienced software developers. Neither is a study of small operations businesses. We cite them for what they show about trials in general: that use falls after the first weeks, and that self-reported time savings can differ from measured ones.

On first-party claims. Statements about "our diagnostic" describe Avolis's own engagements. They are not independent research. The brokerage is an illustrative composite, not a client, and its figures are illustrative.

---

About Avolis Research Group

Avolis Research Group is Avolis's in-house research practice, focused on how operations-heavy small and mid-sized businesses actually adopt AI. It synthesizes primary economic research, government survey data, and results from real implementations into practical, vendor-neutral guidance.

More about Avolis and how we work · Get in touch